Active anemosensing hypothesis: How flying insects could estimate ambient wind direction

Estimating the direction of ambient fluid flow is a crucial step during chemical plume tracking for flying and swimming animals. How animals accomplish this remains an open area of investigation. Recent calcium imaging with tethered flying Drosophila has shown that flies encode the angular direction of multiple sensory modalities in their central complex: orientation, apparent wind (or airspeed) direction, and direction of. Here we describe a general framework for how these three sensory modalities can be integrated over time to provide a continuous estimate of ambient wind direction. After validating our framework using a flying drone, we use simulations to show that ambient wind direction can be most accurately estimated with trajectories characterized by frequent, large magnitude turns. Furthermore, sensory measurements and estimates of their derivatives must be integrated over a period of time that incorporates at least one of these turns. Finally, we discuss approaches that insects might use to simplify the required computations and present a list of testable predictions. Together, our results suggest that ambient flow estimation may be an important driver underlying the zigzagging maneuvers characteristic of plume tracking animals' trajectories.

Methods
This archive contains four datasets described below. For details on data collection please refer to the associated preprint manuscript. For details on data format and processing, please refer to our open source code repository: https://github.com/florisvb/active_anemosensing [1].

Raw data of real-world wind measurements. Data includes wind direction and speed data using two 3-D ultrasonic anemometers (Trisonica mini, Anemoment, Longmont CO) operating at 10 Hz positioned orthogonal to one another to ensure accurate readings of both horizontal and vertical wind components. We collected data in an open desert environment for 120 minutes. Raw data is provided as a collection of binary files for each anemometer. To extract the vertical wind velocity and horizontal wind speed and direction, synchronize the data, and interpolate the two sensors to the same master time base, please refer to the Jupyter notebooks provided here: https://github.com/florisvb/active_anemosensing/tree/main/preprocess_raw_wind_data. Data was collected using custom data logging firmware running on a teensy microcontroller, for details please refer to the following repository: https://github.com/florisvb/gps_wind_station. 

Raw multisensory data from a quadrotor drone flight. Data includes multiple sensor streams from a ~3 minute manually piloted drone flight. The raw data consists of .csv files containing GPS, IMU, 3D ultrasonic anemometer, and 2D optic flow data. To process the raw data, interpolate to a single master time base, and reorient the coordinate frames for consistency, please refer to the Jupyter notebooks provided here: https://github.com/florisvb/active_anemosensing/tree/main/preprocess_raw_botfly_data. For details on the drone hardware and data collection, please refer to our preprint. 

Curated flight trajectory data of freely flying fruit flies. Data includes 700 3D flight trajectories of fruit flies navigating an odor plume in a wind tunnel in an hdf file format. The trajectories were curated from a previously published raw dataset (Pang et al., 2019) and reformated into an hdf format to only include those with durations greater than 1 second during which an odor (ethanol) was encountered at least once while moving (on average) upwind in 0.4 m/s wind. For details on the original data collection refer to van Breugel & Dickinson (2014), and for details on the raw dataset refer to Pang et al. (2019).

Simulation results. Data includes simulation results for ambient wind estimation under four different scenarios described in the associated manuscripts. For each scenario, several different parameter sweeps were done, and an hdf file is provided for each case. See manuscript and associated software (van Breugel, 2022) for details.

References. 

van Breugel, F. Active Anemosensing. (2022). Zenodo, https://doi.org/10.5281/zenodo.6402594.
Pang, Rich et al. (2019). Data from: History dependence in insect flight decisions during odor tracking, Dryad, Dataset, https://doi.org/10.5061/dryad.n0b8m

van Breugel, F. and Dickinson, M. H. (2014). Plume-Tracking behavior of flying Drosophila emerges from a set of distinct sensory-motor reflexes. Current Biology, https://doi.org/10.1016/j.cub.2013.12.023